Data license

The code of Vocab Bloom Hub is released under the MIT license. The dictionary data of the project — the entries, meanings, examples, translations and inflected forms an instance is born with, exported from the card of the dataset, served by the public /api/v1 and published as the HuggingFace dataset Fristail27/vocab-bloom-hub-en — is licensed separately, under the

Creative Commons Attribution 4.0 International (CC BY 4.0), SPDX identifier CC-BY-4.0.

Copyright (c) 2026 Aleksei Ryzhov.

You are free to copy, redistribute and adapt the data for any purpose, including commercially, as long as you credit the source, link to the license and indicate whether you changed anything. A sufficient attribution:

Vocabulary data from the Vocab Bloom Hub English dataset (CC BY 4.0), https://huggingface.co/datasets/Fristail27/vocab-bloom-hub-en

The same terms travel with the data everywhere it goes:

• manifest.json of every export carries license ("CC-BY-4.0"), license_url, attribution and the source, and a LICENSE file lies next to it;
• GET /api/v1/meta returns license, license_url, attribution and source;
• the admin Export dictionary page shows them next to the download;
• the HuggingFace dataset card declares license: cc-by-4.0 and ships LICENSE and NOTICE.

The constant behind them is apps/server/core/constants/data_license.ts: the terms the project's dataset is registered with on an instance.

Datasets of other sources

This license covers the project's own dataset and nothing else. An instance may hold datasets converted from public sources and datasets of its owner's own (docs/datasets.md), and each dataset keeps the license of its source:

Dataset	License
The project's own (default)	CC BY 4.0, this file
English Wiktionary (kaikki.org)	CC BY-SA 4.0 — share-alike
Open English WordNet	CC BY 4.0
Princeton WordNet 3.x	WordNet license
OpenGloss 2.4	CC BY 4.0; additional Princeton WordNet 3.0 terms on words marked as derived from it
CMU Pronouncing Dictionary	BSD 2-Clause, the pronunciations of a WordNet dataset converted with them
A dataset of the owner's own	the license its owner chose: CC0, CC BY, CC BY-SA, CC BY-NC, ODbL or one of their own

While such a dataset is the active one, everything above — the manifest of an export, /api/v1/meta, the word pages — carries its terms, and whoever uses the data of that instance is bound by them. Datasets are read separately, but one word can have several sources and licenses, including the terms of contributions made in a fork. The converters and the code that serves the data stay MIT: the license of a source binds the data, not the program that reads it.

The WordNet license and the one of the CMU Pronouncing Dictionary ask that their notice is kept in full on every copy. For the datasets that carry them the notices are stated in the code (apps/server/core/constants/dataset_notices.ts) and travel as license_text of /api/v1/meta, on the page /dataset-terms of the website and in the LICENSE file of an export. Open English WordNet is derived from Princeton WordNet and carries its notice too.

OpenGloss 2.4 declares CC BY 4.0 in its dataset card. Words marked source: wordnet-3.0 also retain the Princeton WordNet 3.0 notice. The converter records that additional origin only for those words. This release does not declare Wiktionary as a source. Its generated content and source notices remain visible; the CC BY label alone is not the complete set of terms for every OpenGloss word.

Multiple sources and contributions

The dataset's primary license describes its default terms and new contributions. It does not replace the terms of borrowed material. Read a word's origins[], contributions[] and licenses[] together: origin_id associates each license with the source or contribution that supplies its attribution, version, links and notices. license_relation: "all" means cumulative terms; "any" records alternatives explicitly offered by that source.

Creating a fork or copying a word retains the original sources without assigning the receiving dataset's license to unchanged material. An actual content edit in an own dataset adds the contribution terms captured at that time. Changing the dataset's license or version does not rewrite earlier snapshots. Manual transfers can be declared with their sources and licenses; unknown versions stay unknown. These declarations and the software's compatibility checks do not establish rights to material the user does not have permission to reuse.

Keep the complete export, including manifest.json, dataset-format.json, word origins, the history with contribution snapshots, and LICENSE. The marker identifies the format; it does not contain the licenses itself. See sources and licenses and the portable format.

Why CC BY 4.0

MIT is a software license: it is ambiguous for a database of facts and says nothing about attribution of derived datasets. CC BY 4.0 is the standard choice for open lexical data — free for any use, machine-readable, and it asks for exactly the one thing the project needs, credit that leads users back to the source. Share-alike (CC BY-SA) would have kept derived datasets open but discourages embedding the dictionary in products, which is what the public API is for; CC0 would have dropped the attribution the project relies on to be found.

Provenance

Most of the data is generated by language models and is not human-verified; every record says which model produced it. What that means for using the data, the known limitations and how to report errors are in docs/data.md. Pass the notice on: GET /api/v1/meta returns it as notice, the dataset card and the word pages of the website show it.

The models are chosen so that this license can be honoured. A provider whose terms only bind the customer — DeepSeek's Open Platform terms assign the output to the user and expressly allow derivative products and training other models — is compatible with CC BY 4.0; a provider whose terms make the customer prevent third parties from training on the output (xAI's API terms for Grok, as of August 2026) is not, because CC BY 4.0 grants every recipient exactly that right. The published data is generated with models of the first kind; the check is repeated before a new model joins the pipeline. The terms of DeepSeek also ask that the DeepSeek name is used only factually, without logos or any suggestion of endorsement.

Contributions to the data

An entry edited in the admin UI, a correction sent through Report a mistake on a word page and applied by the owner, and a dataset revision contributed through the repository are licensed under the same CC BY 4.0 — the word pages say so next to the form, and CONTRIBUTING.md records it for pull requests.

A correction takes the license of the dataset it corrects. On an instance that serves another dataset, an edit or an applied correction becomes part of that dataset: under CC BY-SA 4.0 for Wiktionary, under the license the dataset is registered with for any other. The form on the word page names the license before the reader sends anything, and the admin UI names it on the card of a word and in every dialog that edits one.

Changes are indicated. CC BY and CC BY-SA ask whoever shares modified material to say that it was modified. An instance keeps a history of edits with every dataset (docs/datasets.md): an entry that was changed or added says so to its readers, what was changed is public with the values before and after, and the history travels with an export — the manifest counts the modified entries. Wiktionary and the WordNet datasets refuse generated content. OpenGloss explicitly declares generated content and accepts it, as do the project's dataset and datasets of the owner's own.

A contributor is named only when they ask. A reader may give a name with a correction and agree to have it published; it is then a part of the history and of its exports, and is removed on request (docs/data.md).

Using data that was changed on an instance

The owner of an instance may edit the data it serves, and the instance says where it did. Whoever takes data from an instance — through the API, from an export, from a word page — takes it with that indication and is asked to keep it:

What the instance gives	Where
Where an entry comes from	source on every word of /api/v1, on every part of one and on every edit of the history
Whether an entry was changed or added	modified on every word of /api/v1 and on every part of one: meanings, forms, translations, links
What was changed, with the values before and after, and who sent the correction	GET /api/v1/words/{word}/history
How many headwords differ from the source	modified_entries of GET /api/v1/meta, and of manifest.json in an export
The history itself, to travel with a copy	vocab-bloom-hub-en-changes.jsonl of an export
The license and the notices of the source	license, license_url, attribution, license_text of /api/v1/meta; LICENSE of an export
The sources and contribution terms of a particular word	origins, contributions, licenses on word responses; origins in exported entry lines and contributions in the history file
The same for every dataset the instance holds, not only the one it serves	the groups of GET /api/v1/words/{word}/datasets, each with its terms; the edits of one through GET /api/v1/words/{word}/datasets/{dataset}/history

What to do with it, under every license a dataset can have:

1. Credit the source: show the attribution of /api/v1/meta and link to the license wherever you show the data, together with the attributions and notices of the word's original sources and contributions.
2. Say that an entry was changed when it was. An entry with modified: true is not, or not only, what its source published: do not present it as the words of Wiktionary, of WordNet or of the project's dataset without saying so. Passing the mark on — a line next to the entry, as the word pages do — is enough; the history is there for whoever wants to know what was changed.
3. Retain inherited terms when adding your own contribution. Editing or forking an entry does not relicense its original material. Contributions take the terms of the dataset where they were made, subject to the source's requirements. For Wiktionary-derived material those include CC BY-SA 4.0's ShareAlike requirement.
4. Keep the notices in full where the license asks for it: the WordNet license and the one of the CMU Pronouncing Dictionary want their text on every copy, a modified one included. It is license_text of /api/v1/meta and the LICENSE file of an export.
5. Pass the indication on with a copy. A dataset you redistribute keeps its manifest.json, its dataset-format.json, its LICENSE, word origins and its complete declared history; an instance that imports them retains the sources, contributions and modification marks.
6. Do not suggest endorsement. The names of the sources and of their authors are used to credit them, not to suggest that they stand behind a changed entry or behind your product.
7. Keep the datasets apart when you read several. GET /api/v1/words/{word}/datasets answers a group per dataset of the instance, and every group states its own license, attribution and license_text. An entry is used under the terms of the group it was taken from: credit each source for its own entries, keep the notices of each, and keep what you build on a share-alike group under that license. Entries of two groups merged into one text are a work under the terms of both.

An entry that was never edited (modified: false) is what the source published, as the converter of the project read it (docs/datasets.md). modified goes back to false when the owner takes a change back or when an update of the dataset replaces the entry with the content of its source.

This is a description of what the licenses ask, written for the users of the software; it is not legal advice, and the text of each license decides.

Your own instance

A self-hosted instance ships the terms of its active dataset in its exports and its /api/v1/meta, and the terms of every dataset it holds with the groups of GET /api/v1/words/{word}/datasets: a dataset is public from the moment it is installed. The terms of the datasets of the catalog — the project's own and the public sources — are stated in the code (apps/server/core/constants/dataset_catalog.ts) and cannot be edited on an instance: what you edit in such a dataset stays under the license of that dataset. Your own words go into a dataset of your own, created empty under a license you choose; its license can be changed later, but a license that was given is not taken back — whoever took the data under it keeps that right, and the change is journaled. A public source that is not in the catalog is added with its converter (apps/server/src/converters/README.md).